numr 0.5.2

High-performance numerical computing with multi-backend GPU acceleration (CPU/CUDA/WebGPU)
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41

//! Unary operation implementation for WebGPU.

use super::helpers::*;
use crate::error::Result;
use crate::runtime::ensure_contiguous;
use crate::runtime::wgpu::shaders::elementwise;
use crate::runtime::wgpu::{WgpuClient, WgpuRuntime};
use crate::tensor::Tensor;

pub(crate) fn native_unary_op(
    client: &WgpuClient,
    op: &'static str,
    a: &Tensor<WgpuRuntime>,
) -> Result<Tensor<WgpuRuntime>> {
    let dtype = a.dtype();
    let a_contig = ensure_contiguous(a);
    let numel = a.numel();

    let out = alloc_output(client, a.shape(), dtype);

    let a_buf = get_tensor_buffer(&a_contig)?;
    let out_buf = get_tensor_buffer(&out)?;

    let params = UnaryParams {
        numel: numel as u32,
    };
    let params_buf = create_params_buffer(client, &params);

    elementwise::launch_unary_op(
        client.pipeline_cache(),
        client.wgpu_queue(),
        op,
        &a_buf,
        &out_buf,
        &params_buf,
        numel,
        dtype,
    )?;

    Ok(out)
}